An improved TCP congestion control algorithm in high speed satellite links

The throughput of traditional TCP degrades drastically in high speed satellite links, where the characteristics of long round trip time (RTT) and high bit error rate (BER) happen together. Two typical improved algorithms, Hybla and Westwood, are designed to solve the problems of long RTT and high BER respectively. TCP-HW which combines the advantages of Hybla and Westwood is proposed in this paper, on the basis of consideration of both long RTT and high BER. At slow start and congestion avoidance phase, TCP-HW adopts a radical congestion window (cwnd) growth policy to eliminate the low data transmission rate bring by long RTT as Hybla does. At fast retransmit and fast recover phase, TCP-HW updates the cwnd according to the estimated available bandwidth in Westwood to avoid the frequently unnecessary reduction of cwnd due to high BER. There, the cwnd can be kept large enough to achieve high throughput. Simulation results in NS2 indicate that the change of RTT and packet error rate (PER) has little impact on TCP-HW and the bandwidth utilization of TCP-HW can reach more than fifty percent in the 1Gbps GEO satellite link with 500ms RTT and 0.001 PER. In addition, TCP-HW still have good fairness.